Physics-Informed Graphical Neural Network for Parameter & State Estimations in Power Systems (2102.06349v1)

Abstract: Parameter Estimation (PE) and State Estimation (SE) are the most wide-spread tasks in the system engineering. They need to be done automatically, fast and frequently, as measurements arrive. Deep Learning (DL) holds the promise of tackling the challenge, however in so far, as PE and SE in power systems is concerned, (a) DL did not win trust of the system operators because of the lack of the physics of electricity based, interpretations and (b) DL remained illusive in the operational regimes were data is scarce. To address this, we present a hybrid scheme which embeds physics modeling of power systems into Graphical Neural Networks (GNN), therefore empowering system operators with a reliable and explainable real-time predictions which can then be used to control the critical infrastructure. To enable progress towards trustworthy DL for PE and SE, we build a physics-informed method, named Power-GNN, which reconstructs physical, thus interpretable, parameters within Effective Power Flow (EPF) models, such as admittances of effective power lines, and NN parameters, representing implicitly unobserved elements of the system. In our experiments, we test the Power-GNN on different realistic power networks, including these with thousands of loads and hundreds of generators. We show that the Power-GNN outperforms vanilla NN scheme unaware of the EPF physics.

• The paper introduces the Power-GNN model that fuses power flow physics with a graphical neural network to enhance state and parameter estimations.
• It employs Kron reduction and is effective under partial observability, outperforming traditional neural network models.
• Experiments using synthetic IEEE test cases demonstrate robust reconstruction of physical parameters and scalable computational efficiency.

The paper "Physics-Informed Graphical Neural Network for Parameter & State Estimations in Power Systems" introduces a novel approach to State Estimation (SE) and Parameter Estimation (PE) in power systems by integrating physical models with Graphical Neural Networks (GNN). The proposed model, named Power-GNN, seeks to address two fundamental challenges in power system management: generating reliable SE predictions and accurate PE, especially under conditions of limited observation.

Key Concepts and Approach:

1. Physics-Informed Machine Learning (PIML): The paper leverages the concept of physics-informed machine learning where the physics of the system is incorporated into the neural network model. This is crucial for models operated by power system operators who require tangible, explainable parameters related to physical characteristics of power systems.
2. Power-GNN Model:
   • The Power-GNN framework embeds the physics of the Electric Power Flow (EPF) into a graphical neural network architecture. This hybrid scheme involves learning both physics-loaded parameters (like admittance matrices) and physics-blind parameters to predict unobserved variables.
   • The model applies the Kron reduction technique for dimensionality reduction, creating an effective, smaller representation of the power network while accurately reflecting the dynamics of the full system.
3. Capability Under Partial Observability:
   • The paper highlights the model's performance in scenarios where only a subset of the system's nodes are visible through real-time measurements, such as providing robust estimates on power flows based on partial observability of phasor measurements.
   • This feature is significant as it addresses the inherent gaps in observation capability of traditional power systems, where real-time data from all nodes might not be available.

Implementation and Findings:

• Data Generation: Synthetic data from IEEE test cases and a PanTaGruEl model were used for simulation to train and test the Power-GNN.
• Results: Experiments showed that Power-GNN outperforms traditional neural network models (vanilla NN) in full and partial observability scenarios. Power-GNN successfully extrapolates and predicts unseen states and demonstrates robust reconstruction of the physical parameters, such as equivalent power line characteristics.
• Computational Efficiency: The training process on large systems illustrated the computational efficiency and scalability of the model. The use of HPC (High-Performance Computing) resources facilitated handling large datasets with extensive network parameters.

Conclusion:

The paper underscores the potential of physics-informed learning in complex systems where traditional machine learning approaches may struggle due to non-standard data distributions and partial observability. By merging physics-based models with machine learning techniques, the Power-GNN provides a tool that delivers both accurate state estimations and reliable parameter characterizations.

Future Work:

The authors suggest expanding this framework to dynamic states and integrating it into broader infrastructure assessments, such as natural gas and district heating systems. This could entail developing further active learning strategies and extending the application of Power-GNN to secure real-time power grid operations.